st: Is Stata inefficient for Matching?

["Ariel Linden, DrPH" <ariel.linden@gmail.com>]

Hi Santosh,

It seems that you are running into all sorts of issues with PSMATCH2.

In this particular case, have you tried running the program on either a
smaller sample, or limiting the nearest neighbour to 1, rather than 5?

I am in compete agreement with Austin Nichols posting from a few days ago
that weighted modeling is the preferred method. Weighting outperforms
matching in all the studies I have seen, and it certainly makes more sense
if you are using such a large data set, looking to match 5 controls to each
case.

That said, if you are adamant about matching, and PSMATCH2 doesn't work for
you, there are other user-written programs in stata for you to consider:
cem, gmatch, fgmatch, nnmatch, optmatch, match, pscore. I would suggest
trying them out and see if you can resolve your issues.

I hope this helps

Ariel   


Date: Wed, 29 Sep 2010 09:30:27 -0400
From: Santosh Kumar <santosh.uh@gmail.com>
Subject: st: Is Stata inefficient for Matching?

Dear Statalist,

I am using -psmatch2 on stata 11 version. My sample is close to 250000
observations. I ran the following code


 psmatch2 improved, pscore(pscore) outcome(diarr) common caliper(.001)
neighbor(5) trim(5)

The issue is it takes forever to show results. The code is running for
last 24 hours with no output. It is computationally intensive. I
wonder if there is an efficient way to do this in less time.
Sometime, two days pass by without any output. KIndly suggest if you
know how to do it efficiently.

Thanks
- -- 

Santosh

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/